Professional Documents
Culture Documents
KEY WORDS: Digital Elevation Model, Vertical Accuracy, Uncertainty, Sea level rise, Vulnerability Assessment, AW3D30,
ASTER GDEM, SRTM DEM, Mindanao, Philippines
ABSTRACT:
We assessed the vertical accuracies and uncertainties of three freely-available global DEMs as inputs to elevation-based sea-level
rise vulnerability assessment of Mindanao, Philippines - an area where above average SLR of 14.7 mm/year was recently found.
These DEMs are the Shuttle Radar Topography Mission (SRTM) DEM, ASTER Global DEM (GDEM Version 2), and ALOS
World 3D-30 (AW3D30). Using 2,076 ground control points, we computed each DEM’s vertical accuracies and uncertainties, and
from these we determined the smallest increment of sea-level rise (SLRImin) that should be considered when using the DEMs for
SLR impact assessment, as well as the Minimum Planning Timeline (TLmin) for an elevation-based SLR assessment. Results of
vertical accuracy assessment revealed Root Mean Square Errors of 9.80 m for ASTER GDEM V2, 5.16 m for SRTM DEM, and
4.32 m for AW3D30. Vertical uncertainties in terms of the Linear Error at 95% Confidence (LE95) were found to be as follows:
19.21 m for ASTER GDEM V2, 10.12 m for SRTM DEM, and 8.47 m for AW3D30. From these, we found that ASTER GDEM2 is
suitable to model SLR increments of at least 38.41 m and it will take 2,613 years for the cumulative water level increase of 14.7
mm/year to reach the minimum SLR increment afforded by this DEM. For the SRTM DEM, SLRImin and TLmin were computed as
20.24 m and 1,377 years, respectively. For the AW3D30, SLRImin and TLmin were computed as 16.92 m and 1,151 years,
respectively. These results suggest that the readily available global DEMs' suitability for mapping coastal inundations due to SLR in
our study area is limited by their low vertical accuracies and high uncertainties. All the three DEMs do not have the necessary
accuracy and minimum uncertainties that will make them suitable for mapping inundations of Mindanao at smaller increments of
SLR (e.g., SLR ≤ 5m). Hence, users who apply any of these DEMs for SLR impact assessment at SLRIs lower than the DEM’s
SLRImin must be cautious in reporting the areas of SLR vulnerable zones. Reporting the inundated areas as a range instead of a
singular value for a given SLR scenario can highlight the inherent accuracy and uncertainty of the DEM used in the assessment.
* Corresponding author
where zdata I is the DEM elevation of the Ith check point, zcheck I is
the true (ground) elevation of the Ith check point, n is the
number of points being checked, and I is an integer from 1 to n.
Given the annual SLR rate and a DEM’s RMSEz, TLmin would
provide the user of that DEM the number of years for the
cumulative water-level increase to reach the minimum sea-level
rise increment (SLRImin) afforded by the elevation data (Gesch,
2013).
Figure 1. Map showing the location of Mindanao Island, Similar to Gesch (2013)’s illustration, a DEM with RMSEz of
Philippines and the provinces within. 38.3 cm would allow modeling of only 1.5 m or more of SLR.
Using an increment smaller than 1.5 m, according to Gesch
(2013) “will give questionable results as the increment will not
2. METHODS AND MATERIALS be within the bounds of the statistical uncertainty of the given
elevation data”. Assuming the annual SLR rate is 10 mm/year,
2.1 Determining a DEM’s Vertical Uncertainty using the TLmin can be estimated as 150 years which means that it will
Gesch (2013) Approach take 150 years for the cumulative water-level to reach the
minimum SLR interval afforded by the elevation data.
Gesch (2013) demonstrated how to take a DEM's vertical
uncertainty into consideration in elevation-based SLR 2.2 30-m Global DEMs Used
vulnerability analysis by quantifying two specific parameters
namely, the Minimum Sea Level Rise Increment (SLRImin; also The AW3D30 Version 1 DEM tiles of Mindanao (Figure 2)
referred to as SLR Interval) and the planning horizon which the were downloaded from
author referred to as the Minimum Planning Timeline or TLmin. http://www.eorc.jaxa.jp/ALOS/en/aw3d30/. These tiles were
generated using stereo images acquired by the Panchromatic
According to Gesch (2009), a DEM-based SLR assessment can Remote Sensing Instrument for Stereo Mapping (PRISM)
use small intervals of water-level change as long as the DEM’s sensor onboard the Advanced Land Observing Satellite (ALOS)
vertical accuracy truly support these intervals, i.e., the intervals from 2006 through 2011(Takaku et al., 2014).
must be within the bounds of the statistical uncertainty of the
elevation data. This minimum interval, SLRImin, can be Each tile of the DEM was provided in two types: AVE and
determined using the equation (Gesch, 2013): MED according to the method used when resampling from the
5-meter mesh version (AVE = average; MED = median). For Vertical accuracies of the three DEMs were assessed by
this study, we used the AVE tiles. All the tiles were mosaicked comparing the DEM elevations with those of the GCPs. Two
and saved in GeoTIFF format using Global Mapper software, types of assessments was conducted. The first assessment
and reprojected from WGS 1984 geographic coordinates system considered all the GCPs regardless of the order of their
to Universal Transverse Mercator Zone 51 projection (retaining accuracy, while the second assessment look into the effects of
WGS 1984 as its horizontal datum) using ArcGIS 9.3 software. the different orders of accuracy of the GCP on the computed
The elevation values in the AW3D30 are considered “height DEM vertical accuracies and uncertainties.
above sea level” with the Earth Gravitational Model 1996
(EGM96) as the vertical datum (JAXA, 2017). Missing data due
to cloud cover is evident in the AW3D30 DEM Version 1
(shown as white gaps in Figure 2). However, in the most recent
version (Version 1.1; March 2017), these gaps or voids have
been complemented by existing DEMs.
H=h–N (6)
The DEM elevations at each point were extracted using ArcGIS maximum and minimum elevation of areas that can be affected
9.3 software. The differences in elevation, which are the by a SLR of 1.23 m at 95% confidence level.
measured errors in the DEMs, were computed by subtracting
the GCP elevation from its corresponding DEM elevations.
Positive errors represent locations where the DEM was above 3. RESULTS AND DISCUSSION
the GCP elevation, and negative errors occur at locations where
the DEM was below the control point elevation. From these 3.1 Vertical Accuracies of the DEMs Using All GCPs
measured errors, the mean error and RMSEz for each DEM
were calculated. The mean error (or bias) indicates if a DEM The calculated errors of the DEMs plotted with the EGM96
has an overall vertical offset (either positive or negative) from elevation of all the GCPs are shown in Figure 4, while the
the true ground level (Gesch et al., 2012). summary of computed error statistics are listed in Table 2. The
error plots indicates that there is no clear relationship between
The values of LE95, SLRImin and TLmin were then computed the calculated errors and elevation for all DEMs. Another
from the RMSEz of each DEM. observation is that there is no preference for positive or
negative errors as the plotted DEM errors are uniformly
distributed on both sides of the zero error axis. This finding is
similar to that of Gesch et al. (2012) in their vertical accuracy
assessment of ASTER GDEM2.
Tadono, T., Nagai, H., Ishida, H., Oda, F., Naito, S., Minakawa,
K., Iwamoto H., 2016. Generation of the 30 m-mesh global
digital surface model by ALOS PRISM. In: The International
Archives of the Photogrammetry, Remote Sensing and Spatial
Information Sciences, Vol. XLI-B4, pp. 157-162.